Red Flags to Watch for in an AI Visibility Proposal

A guarantee is not a deliverable

A deliverable is something the agency can prove it completed. A dated baseline, a correction to a business profile, a rewritten service page, a source review and a monthly change log all fit that description. Being named by ChatGPT next Tuesday does not. Neither does a promised number of leads.

OpenAI says ChatGPT Search results and citations can be incomplete, outdated or wrong, and tells users to inspect the sources behind an answer. It does not publish an organic position that an outside agency can reserve. Generative models are not static, predictable slots where the same output will be generated every time. Google makes a similar distinction in its own guidance for hiring an SEO. It warns businesses about guaranteed rankings, secret relationships and priority-submission claims, including advice sold for AI search experiences.

The Federal Trade Commission's test is useful here even before a dispute exists. Advertising claims include what a reasonable buyer would take from the whole pitch, not only the exact words on the page, and a business needs support for the claims it makes before it runs the ad. A money-back guarantee does not substitute evidence that a performance will be achieved.

Read every outcome promise beside the work that is supposed to produce it. If the work can be inspected but the outcome cannot be controlled, the contract should guarantee the work and describe the outcome as something to observe.

A dashboard needs a denominator

A percentage can look precise while hiding most of the test. Ask which customer questions were run, which assistant and mode answered them, what location was assumed, how many times each question was repeated, which competitors counted and how the software turned the raw answers into the number on the page.

One favorable screenshot cannot answer those questions. It may record a real answer, but it does not show whether the result repeats or whether a slightly different customer question produces a different list. The AI visibility score guide explains how a score can be mathematically valid and still be commercially unhelpful.

A proposal does not need a giant research program. It does need a measurement rule both sides can read. The first report should preserve the starting questions and raw answers. Later reports should use the same conditions unless a change is disclosed. If a question is added, removed or rewritten, the report should say so rather than drawing a clean trend line across two different tests.

The same caution applies to case studies. Ask to see the starting point, the actual work, the observation window and what remains uncertain. A before-and-after screenshot can illustrate a result. It cannot establish that the agency caused every change between them.

Old SEO work does not become new work because the label changed

AI visibility work often overlaps with ordinary SEO. That is not a flaw. Search and AI products both need public pages they can access and understand, accurate business information and real evidence about the company. The overlap becomes a billing problem when the proposal charges a second time for the same page edits, profile maintenance, technical repairs or reporting without naming what is new.

Compare the proposal with the current SEO agreement line by line. If both include service-page improvements, internal links, Business Profile work and content production, ask which provider owns each task. The AI visibility proposal should identify any additional work, such as a separate question set, multi-product answer checks, source analysis, factual accuracy review or a different reporting record.

Be more careful when the plan calls for hundreds of pages, paid mentions, forum posts or reviews. Google says its spam policies apply to attempts to manipulate both ordinary results and generative AI responses. It specifically prohibits scaled content created mainly to manipulate rankings rather than help people. A proposal should name where third-party work will appear, which platform or editorial rules apply, who approves the claim and what happens if the placement is removed.

Paying an agency buys professional work, not a paid organic recommendation inside ChatGPT. The scope should make that distinction impossible to miss.

Your accounts should still be yours

Account ownership can look like a small administrative detail until an agency relationship ends. Then it decides whether the business still controls its domain, website, analytics, Search Console, Business Profile, ad history, original content and measurement records.

Google's Business Profile rules are direct: the client must retain ownership or co-ownership, and a third party must provide a quick way to end the relationship and relinquish its permissions. That policy governs Business Profile management. It is also a sensible standard to apply when negotiating the rest of the engagement.

The proposal should name every account, its owner during the project and the files the client receives when the work ends. That includes the domain, website CMS or repository, analytics, Search Console, Business Profile, ad accounts, purchased content, raw exports and any paid workspace that the contract says will transfer. It should also state what access the agency receives and how quickly that access will be removed.

Do not accept shared passwords as a substitute for ownership. Create the business account, keep the recovery information under business control and invite the agency with the permissions it needs. The agency can still do the work. You avoid having to recover your own property from a former supplier.

The exit terms belong in the proposal

A contract can have a clear cancellation date and still leave the practical exit unanswered. What happens to the website changes, research notes, question set, raw answers, correction history, content drafts and reporting data? Which subscriptions stop? Which files are delivered? How long does the transfer take?

This is where many otherwise detailed proposals become vague. Baselines and monthly measurement are easy to sell because they make the service look concrete. Ownership and exit terms are less exciting, but they determine whether the business keeps any useful record after the final invoice.

Ask for a plain exit package. It should contain the assets the business paid to create, an export of the measurements needed to understand past reports, a current list of account access, and a change log showing what was altered. It should also state what cannot transfer, such as a vendor's proprietary software, before that limitation becomes a surprise.

A strong agency should be willing to recommend stopping. If the baseline shows that customers are unlikely to use these tools for the business, or that ordinary website and profile work is the only sensible next step, the proposal should not force a long retainer merely to keep a dashboard alive.

Read each promise as a contract line

Before signing, make the agency answer these questions in the proposal or statement of work:

Vague wording is not made safer by a polished dashboard. Ask the seller to rewrite broad promises as deliverables, measurement conditions, ownership and exit terms. If that makes the offer sound smaller, it is probably becoming more honest.

• What exact work will be delivered, and what outcome will only be measured?
• Which questions, products, locations, repetitions and counting rules create the baseline?
• Which tasks overlap with the business's current SEO, website or reputation work?
• Which pages, listings, reviews, mentions or third-party claims require the owner's approval?
• Which accounts, files, content and raw records remain with the business?
• What is delivered when the contract ends, and how quickly is access transferred or removed?
• What will the agency recommend if the baseline finds no worthwhile opportunity?